A machine learning approach for adhesion forecasting of cold-sprayed coatings on polymer-based substrates

Cold spray is a novel production technology for creating metallic layers on various materials. Using a pressurized gas travelling at supersonic speeds, the metallic particles are accelerated and impact the target surface obtaining adhesion through mechanical interlocking between the powders and the substrate. This method is especially well suited for coating thermosensitive materials like composites since it only requires a little amount of heat, as the powders remain in a solid state. The quality and comprehension of this manufacturing process can be greatly improved by using machine learning techniques. In order to evaluate the characteristics of the particle's deformation upon collision, the goal of this work is to forecast it using machine learning approaches. The parameters chosen as an input for the model were related to 3 macrocategories: process parameters, powder parameters and substrate parameters. As regards the output parameters, flattening and penetration were chosen as they are the main characteristics of the coating on which homogeneity and adhesion depend. In order to obtain reliable results, a mix of data FEM and experimental data were used to train the neural network. The model was then tested on a dataset of experimental data.